Most existing mobility-assisted data access techniques in delay tolerant mobile ad hoc networks (DT-MANETs) are designed to disseminate data to one or several particular destinations. Different from these works, we study the data diffusion problem which diffuses data among all moving nodes so that the nodes that are interested in this data item can get it easily either from their encountered friend nodes or stranger nodes. To reduce the data access delay, we introduce four socialaware data diffusion schemes based on the social relationship and data similarity of the contacts. We also provide solutions to quantify data/interest similarity and to determine whether two nodes are friends or strangers. Theoretical models are developed to analyze the data diffusion process and compare the performance of the four proposed diffusion schemes in terms of diffusion speed and query delay. We use real traces of human contacts to emulate data diffusion under different schemes. Both theoretical analysis and experimental results imply an interesting fact: to achieve better diffusion performance, each node should first diffuse the data similar to their common interests when it meets a friend, and first diffuse the data different to their common interests when it meets a stranger.